Generalization
Our model works here. Will it work somewhere else?
New patients, hospitals, devices, and acquisition protocols can change the problem.
AI/ML FOR MEDICAL DEVICES
From promising model to validation-ready algorithm.
I help medical-device teams solve the difficult AI problems between prototype and clinical validation.
Model performance. Signal quality. Generalization.
The technical judgment to know what comes next.
Principal-led consulting · Deep learning · Biosignals
SYSTEM / GENERALIZATION
01Development data
A promising resultAI / ML model↓
02 / TEST WHAT CHANGES
New patients?
New sites?
New devices?
03
Clinical validationEvidence for the intended use
15+ years
Building applied ML systems
Medical AI
Research → production → validation
Biosignals
Physiology, ultrasound & diagnostic AI
4,000+
Professionals mentored through MIT & Johns Hopkins programs
01 / THE TECHNICAL GAP
The difficult part starts
after the prototype works.
Strong development metrics can leave important questions unanswered. Real patients and new conditions reveal what the average score cannot.
Clinical performance
It performs well on average. But fails where it matters.
Subgroups, difficult signals, and clinically important ranges need their own evaluation.
Model diagnostics
We keep changing the model. Performance isn’t improving.
Is the limitation architecture, signal, data, labels, or the way the problem is framed?
Validation readiness
We have metrics. What does “good enough” mean?
Connect acceptance criteria and patient-level analysis to the decision before model lock.
Technical leadership
We need senior ML judgment.
Prioritize experiments, challenge assumptions, and guide the team through a difficult decision.
We work in the gap between “the model performs well” and “we have strong evidence it works for its intended use.”
02 / DEFINED ENGAGEMENTS
A clear question.
A concrete way forward.
Fixed scope. Named deliverables. A defined timeline and fee.
Work directly with the principal from start to finish.
01 / READINESS7–10 business days
AI/ML Algorithm Readiness Review
Is the algorithm ready for clinical validation?
An independent technical review of your model, data, evaluation strategy, and remaining risks before model lock.
WHAT YOU RECEIVE
- Readiness assessment & technical gap matrix
- Risk register & priority experiment plan
- Model-lock recommendation
- 60–90 day roadmap & executive readout
Technical review areas
Data independence and leakage; ground truth; patient-level performance; subgroups; site, device, and protocol generalization; model stability; and remaining experiments.
US$7,500USD · fixed-scope engagement
View scope and deliverables02 / DIAGNOSIS3–5 weeks
Medical AI Model Diagnostic & Rescue Sprint
Why isn’t the model working?
A focused investigation when performance plateaus, external validation disappoints, or the bottleneck remains unclear.
FOCUSED INVESTIGATION
- Error stratification & signal-quality analysis
- Data, labels & generalization investigation
- Architecture review
- Diagnostic experiments & next-step recommendations
From US$15,000USD · fixed-scope engagement
03 / VALIDATION2–3 weeks
AI/ML Validation Strategy Sprint
How should we prove the algorithm works?
A technical evaluation framework designed around your algorithm, its intended population, and the product decision.
WHAT YOU RECEIVE
- Evaluation strategy & acceptance criteria
- Metrics & subgroup analysis
- Generalization & failure-analysis plan
- Technical validation test matrix
From US$12,500USD · fixed-scope engagement
Need implementation afterward? Model remediation, new experiments, deep-learning architecture work, clinical-data analysis, and fractional AI/ML leadership can be scoped separately.
03 / THE APPROACH
Understand the model.
And what happens beyond it.
Another architecture is not always the answer.
The evaluation starts with the actual system: what information the signal contains, how reliable the ground truth is, which errors matter clinically, and what changes outside the development environment.
I help isolate the uncertainty before recommending more modeling work. You get an evaluation grounded in your algorithm’s behavior and the decision your team needs to make.
FIND THE LIMITING FACTOR
SignalDataLabelsEvaluationArchitecture
ONE CONNECTED TECHNICAL PICTURE
01Signal & data
02Deep learning
03Model behavior
04Patient-level performance
05Generalization
06Clinical validation
04 / PRINCIPAL-LED, BY DESIGN
The person you meet
is the person doing the work.
I’m Cristiano Aguiar, Ph.D., founder of CSA Intelligence.
I’ve spent more than 15 years building machine-learning systems across engineering and healthcare, with recent work focused on medical-device AI, deep learning, physiological signals, and clinical validation.
My experience spans early research, deployed algorithms, difficult model-performance problems, and the technical evidence needed in regulated product development.
Alongside consulting, I mentor professionals in AI programs from MIT Professional Education and Johns Hopkins University, delivered in collaboration with Great Learning.
Research, invention, and mentoring
Co-author of peer-reviewed sleep research on deep learning and physiological signals.
- Sleep-apnea detection with deep neural networks Nature and Science of Sleep · 2023
- Respiratory motion and airflow estimation during sleep Nature and Science of Sleep · 2022
- US patent 12,678,095 · Breathing-signal analysis Named co-inventor · Issued July 2026
- Mentor biography at Great Learning 15+ years in applied AI · 4,000+ professionals mentored
START WITH THE TECHNICAL QUESTION
01
Fit call
Understand the algorithm and the decision ahead.
02
Fixed scope
Agree the inputs, deliverables, timeline, and fee.
03
Technical work
Investigate the model, data, and evidence.
04
Clear decision
Receive findings, priorities, and recommendations.
LET’S FIND THE NEXT STEP
What does your model
need to prove next?
Approaching clinical validation—or facing a model that isn’t behaving the way it should? Tell me where the algorithm is today and what decision your team needs to make.
30-minute qualification call
Remote · Canada / US / international